Five-Stage AI Article Pipeline with Human Gates

Hand-drawn sketch of the five-stage article pipeline: research, outline, draft, score and polish, each a Make.com scenario with an editor checkpoint in Airtable between stages, and a single stage re-run loop instead of regenerating the whole article

An editorial team that had rejected one-shot AI writing now runs a five-stage pipeline on Make.com, Airtable and OpenAI with a human gate between every stage. Articles of 1,800 to 4,000 words ship with a source list, and any single stage can be re-run on its own. The client reports production time per article fell from six to nine hours to sixty to ninety minutes; that figure is the client's estimate, not a measured number.

Built by Prem Patel, Make Level 5 Expert and Zapier Certified Professional; Nex Automations is listed in both official partner directories. This is one of the 1,200+ systems we have shipped, and the pattern we reuse whenever an editor has to trust the output.

The situation

Most AI content pipelines fail at the same point: the editor stops trusting them and goes back to doing it by hand. The team had tried one-shot AI writing and rejected it: no sources, no way to fix one paragraph without regenerating everything, and no visibility into why the model wrote what it wrote.

This one was designed around the editor from the start.

The 3 problems

  1. No sources. A draft with no research trail cannot be checked, so every claim had to be re-verified by hand.
  2. All-or-nothing regeneration. Fixing one weak section meant regenerating the whole article and losing the parts that were fine.
  3. No visibility. Editors could not see what the model was given, so they could not correct the brief instead of the output.

What was built

Five stages, each a Make.com scenario, each with a human checkpoint in Airtable: research, outline, draft, score, polish.

How it flows

  1. An editor creates the brief in Airtable; setting the status triggers the research scenario, which returns sources and notes
  2. Outline stage drafts the structure from the research; the editor approves or edits it in Airtable
  3. Draft stage writes section by section against the approved outline
  4. Score stage grades the draft against the brief for readability, coverage and fit before a human reads it
  5. Polish stage produces the final copy plus metadata: SEO title, description, headline variants and social versions
  6. Ten background automations handle the plumbing: Slack notices, source checks, formatting and hand-off to the CMS

Airtable is the control panel. A piece cannot move to the next stage until a person sets its status. Every stage stores what the model was given and what it returned, so an editor can see why a draft says what it says. Notes carry forward from one stage to the next, and if a draft is wrong, the editor re-runs that one stage, never the whole article.

Tools used

Results

MeasureValue
Article length1,800 to 4,000 words, with a source list per piece
Re-run granularityOne stage at a time
Background automations10
Production time per article (client-reported)6 to 9 hours down to 60 to 90 minutes
Who runs itThe editorial team, without an engineer

Honest limits

Related guides

FAQ

Q: How do you build an AI content pipeline editors will actually use? A: Split it into stages with a human gate between each, keep the state in Airtable so the editor sees inputs and outputs, and let them re-run one stage. That is the five-stage design here.

Q: Can Make.com orchestrate a multi-step AI writing workflow? A: Yes. Each stage is a scenario triggered by an Airtable status change, so the workflow is visible, pausable and editable by the team.

Q: Why not generate the whole article in one prompt? A: Because one weak section then costs the whole piece, and nobody can see why the model wrote what it wrote. Stage outputs are stored and re-runnable, which is what made the editors trust it.

Q: What does a five-stage AI article pipeline cost to run? A: Model usage per article is small compared with editor time; the main cost is the Make.com operations across five scenarios plus the background automations. We size the plan from expected articles per month on a scoping call.

Want content automation your editors trust?

If your team has rejected one-shot AI writing, the fix is the workflow, not a bigger model.

Book a Call →